Using Gravel, we can distribute six applications, each with frequent small messages, across a cluster of eight GPU-accelerated nodes. Compared to one node, ...

Abstract: Distributed systems incorporate GPUs because they provide massive parallelism in an energy-efficient manner. Unfortunately, existing programming ...

Gravel: Fine-Grain GPU-Initiated Network Messages. SC'17, November 2017, Denver, Colorado USA. 5 messages from the GPU's data-parallel hardware to the NI.

Gravel leverages diverged work-group-level semantics to amortize synchronization across the GPU's data-parallel lanes to enable fine-grain PGAS-style ...

Using Gravel, we can distribute six applications, each with frequent small messages, across a cluster of eight GPU-accelerated nodes. Compared to one node, ...

Bibliographic details on Gravel: fine-grain GPU-initiated network messages.

Gravel: fine-grain GPU-initiated network messages. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and ...

Gravel: fine-grain GPU-initiated network messages · Author Picture Marc S. Orr ... gem5-gpu: A Heterogeneous CPU-GPU Simulator · Author Picture Jason Power.

Co-authors ; Gravel: Fine-grain gpu-initiated network messages. MS Orr, S Che, BM Beckmann, M Oskin, SK Reinhardt, DA Wood. Proceedings of the International ...

We accomplish this by exploring the GPU's coarse-grained memory model and correcting semantic mismatches when GPUs wish to directly interact with the network.